Everrest weekly alert digest

Week ending 30 June 2026 · 15 series · detector D4 robust 3-sigma · seed 42 · all money is GMV unless labelled platform revenue (Everrest takes 11%)

$104,558
GMV, last 7 days
$11,501 platform revenue
55%
percentile against the last 52 weeks
typical week $102,600, usual swing 14%
0
alerts this week
11 in the last 28 days
15
series watched
1 total, 8 categories, 6 regions
Read this before trusting anything below. This rule finds 23% of planted changes at 8.2% precision. Alerting at random on this bench would score 7.6%, so its lift over guessing is 1.08x. It arrives a median 14 days after a change starts and produces about 58 false alarms a year. A lift near 1.0 means this rule has not demonstrated that it beats guessing on precision. It is shown because measuring that is the point of the exercise, not because the alerts are reliable.

Alerts this week

24 Jun to 30 Jun 2026 · materiality floor $2,000 of GMV

0 alerts this week. Nothing crossed the threshold. That is the result, not an omission - and on this board it is the usual result.

Largest sub-threshold moves

24 Jun to 30 Jun 2026 · these are NOT alerts - none crossed the threshold

seriesdayactual expectedgap (GMV)z
cat: Sports & Outdoor26 Jun$618$1,392-$773-2.7
cat: Grocery30 Jun$6,691$1,730$4,961+2.4
region: MY25 Jun$4,756$2,007$2,749+2.3
cat: Pet Supplies27 Jun$893$2,370-$1,477-2.1
cat: Apparel29 Jun$712$1,710-$998-2.1
cat: Toys & Kids26 Jun$628$1,876-$1,248-2.0

What moved the total

October 2025 vs November 2025 · reference analysis, not this week

slicebeforeafter changeshare of move
Grocery / ID$12,242$47,314$35,0717%
Grocery / VN$3,800$32,499$28,6996%
Apparel / SG$15,486$40,536$25,0505%
Grocery / SG$21,713$42,246$20,5334%
Beauty / SG$12,570$30,562$17,9924%
Electronics / SG$12,031$29,485$17,4544%
Segment changes sum to the total change exactly, asserted in code. A share above 100 percent means that slice moved further than the total and was partly offset by slices moving the other way.

Average order value: mix or rate

October 2025 vs November 2025 · reference analysis, not this week

Observed $-0.99 = mix $+0.83 + rate $-1.82.

The two halves point in OPPOSITE directions. The mix term ($1) worked in your favour and was more than cancelled by the rate term (-$2), so the headline fell because of what each segment charges. The largest single contributor is Beauty. Reading this as a mix story would send someone to the wrong place.
categoryAOV beforeAOV after own changeshare change mix effectrate effect
Beauty$141.68$112.63$-29.04-1.2%$-0.03$-3.88
Apparel$137.84$132.19$-5.65+0.7%$-0.01$-0.79
Home & Living$140.95$134.82$-6.13-0.1%$-0.00$-0.73
Sports & Outdoor$126.53$127.44$+0.91-0.2%$+0.02$+0.09
Pet Supplies$130.83$134.21$+3.38-1.1%$+0.09$+0.35
Toys & Kids$117.01$123.17$+6.16-0.1%$+0.02$+0.86
Electronics$115.10$122.79$+7.70+0.6%$-0.14$+0.96
Grocery$204.34$213.87$+9.52+1.4%$+0.88$+1.32

Detector bench

20 simulated replicates · 12 planted changes each · base rate 7.6%

rungrecallprecision lift vs chancemedian lag (days) false alarms per year
D1 week-over-week 15pct98%7.5%0.99x0.04,121
D2 3-sigma on level48%7.4%0.98x10.0145
D3 3-sigma on residual23%10.2%1.35x14.033
D4 robust 3-sigma23%8.2%1.08x14.058
D5 calibrated null6%9.6%1.27x21.010
D6 calibrated + FDR7%8.4%1.11x21.013
D7 FDR + persistence0%0.0%0.00xnan1
Measured against changes planted at known times and sizes. Recall cannot be computed on production data, because the misses are by definition absent from the alert log. Lift is precision divided by the base rate; 1.0x is indistinguishable from alerting at random.

What this digest does not do

No severity tiers, no ownership, no runbooks, no suppression and no grouping. The board is nested - TOTAL is the sum of the categories and also of the regions - so one category incident produces at least three alerts, and nothing here collapses them into one. Every count above is an alert count, never an incident count.